The Research And Implementation Of The Multi-organ Segmentation System For Chest Based On 3D V-net

According to the data shows that malignant tumor(also known as cancer)has been one of the causes of harm to people’s health,and its rate accounts for a part of the death factor.Every year in China,the cost of treating malignant tumor is very large,and malignant tumor is on the rise in China.Radiotherapy is a relatively effective method to deal with malignant tumors.The method requires a high degree of radiation dose accuracy,the success rate of radiotherapy,and the cell growth before and after treatment of tumor-affected organs and adjacent tumor-affected organs(known as dangerous organs(OARs)).The radiotherapy program begins with the screening of computed tomography(CT).In order to better specify the radiotherapy plan,the target tumor and the organs located near the target tumor(OARs)need to be described.It is necessary to segment the organs in the CT image.The gold standard of segmentation is mainly based on doctors’ manual demarcation,combined with doctors’ anatomical knowledge of organs.However,medical image data is especially large.There are hundreds of slices in a single case.It is time-consuming for doctors to label each slice,and the results can vary with different operators.Therefore,the scheme of automatic segmentation algorithm is proposed to accurately segment the organs around the malignant tumor,which is helpful to explain the changes of the inherent position and morphology in the patient,so as to promote the adaptability and computer-assisted radiotherapy.It is not only helpful for the quantitative evaluation of the region of interest,but also for accurate diagnosis,prognosis prediction,surgical planning and intraoperative guidance.Many neural network models based on image segmentation have been proposed and good segmentation results have been obtained.Due to the particularity of medical images,it was not until 2015 that u-net network model was launched.As the first segmentation network for medical images,the medical image segmentation results have a good accuracy according to its advantages.In biomedical field,there are a lot of 3d data,and it is unrealistic to label the data layer by layer by layer into 2d data.Moreover,it is inefficient and highly possible to over-fit the training with all the data of the whole 3d volume.The modified 3d u-net can well solve the problem of 3d data input and training.The 3d v-net network model adds residual modules and different loss functions into the structure of the 3d u-net network model.In this paper,the 3d v-net segmentation network is mainly used to realize the automatic segmentation of multiple chest organs on the public data set provided by the SegTHOR challenge,and the system is designed to visualize the region of interest by combining with the automatic segmentation algorithm.We carried out normalization,resampling and other preprocessing on the public data sets,made training sets and test sets,and obtained multi-organ segmentation results with better segmentation accuracy by importing the network model for training.The interface design for visualization USES PyQt5,OpenGL,and nibabel graphics processing libraries in a python3.6-based environment,mainly including the login and main interface.The functions include user account security,file reading and writing,selection of parameters,training network model,and visualization of two-dimensional and three-dimensional medical images.In this paper,the system design is combined with the deep learning algorithm of medical image.Based on 3D v-net network,the chest multi-organ segmentation system can quickly and automatically extract the region of interest,and assist the doctor to evaluate the region of interest in the radiotherapy plan.